Спросить
Войти

ИССЛЕДОВАНИЕ МЕТОДОВ СЕГМЕНТАЦИИ ДЛЯ РАСПОЗНАВАНИЯ ГРАНУЛ ИОНООБМЕННЫХ СМОЛ

Автор: Ибадуллаев Кудрат Кувондик Угли

ТЕХНИЧЕСКИЕ НАУКИ

STUDY OF SEGMENTATION METHODS FOR RECOGNITION OF

ION-EXCHANGE RESINS GRANULES Ibadullaev K.K.1, Rusak A.V.2 Email: Ibadullaev689@scientifictext.ru

1Ibadullaev Kudrat Kuvondik ugli - Master; 2Rusak Alena Viktorovna - Candidate of Technical Science, Docent, DEPARTMENT OF SOFTWARE ENGINEERING AND COMPUTER SYSTEMS, SAINT-PETERSBURG NATIONAL RESEARCH UNIVERSITY OF INFORMATION TECHNOLOGIES, MECHANICS AND OPTICS, ST. -PETERSBURG, SAINT-PETERSBURG

Abstract: in this paper the methods of segmentation of images of ion-exchange resins are study. This problem arises when solving the problem of automating the process of determining the osmotic stability of grains of ion-exchange resins of grains of ion-exchange resins and is a preliminary step for the classification of grains of ion-exchange resins. Standard methods of computer vision provide an acceptable result for monodisperse resin. However, to solve the problem of recognition of ordinary ion-exchange resins, obviously, the use of other approaches, in particular, convolutional neural networks, is required. Keywords: ion-exchange resin, image segmentation, image filtering, Canny edge detector, Hough transform, watershed with markers.

ИССЛЕДОВАНИЕ МЕТОДОВ СЕГМЕНТАЦИИ ДЛЯ РАСПОЗНАВАНИЯ ГРАНУЛ ИОНООБМЕННЫХ СМОЛ Ибадуллаев К.К.1, Русак А.В.2

1Ибадуллаев Кудрат Кувондик угли - магистр; 2Русак Алена Викторовна - кандидат технических наук, доцент, факультет программной инженерии и компьютерной техники, Санкт-Петербургский национальный исследовательский университет информационных технологий, механики и оптики, г. Санкт-Петербург

Аннотация: в данной статье рассматриваются методы сегментации изображений ионообменных смол. Эта проблема возникает при решении задачи автоматизации процесса определения осмотической устойчивости зерен ионообменных смол и является преддварительным этапом для классификации зерен ионообменных смол. Стандартные методы компьютерного зрения обеспечивают приемлемый результат для монодисперсной смолы. Однако для решения проблемы распознавания обычных ионообменных смол, очевидно, требуется использование других подходов, в частности, сверточных нейронных сетей.

UDC 004.93

Problem statement. Ion exchange resins (ion-exchangers) are widely used in different fields of industry. Most often they are used for water softening and water purification. Ion exchange resins are solid polymers in the form of small microbeads (radius 0.25-0.5 mm) usually yellowish or white. One of the main characteristics of ion exchange resins is osmotic stability. Currently, resins quality control is performed manually by counting the percentage of the number of unbroken beads (whole and cracked) to the total number of beads (grains and fragments) when studying them with help a microscope or a photo-magnifier. This is a

rather laborious process. To increase the efficiency of resins quality control, automation of this process is required. The solution of this problem requires the development of a method for recognition of ion exchange resin beads in the image and their classification into whole and cracked [1].

The initial data are images of ion-exchange resins beads obtained using an optical microscope. An example of the initial data with resin samples is shown in figure 1. The beads of monodisperse ion-exchange resins (Fig. 1a) are quite uniform in size and color. The difficulty of recognizing beads of ordinary ion-exchange resins (Fig. 1b) determined by difference of particle sizes and shapes, color inhomogeneity, partial overlap of the samples, blurriness at the edges of the image, coincidence of the background color and the color of the particles. In addition, the original images from the microscope contain an arbitrary number of objects.

Thus, the first step in solving the problem of resin beads recognition is pre-processing and segmentation of the image to obtain images of individual resin samples for their subsequent classification into whole and cracked. In addition, the obtained images can be used to increase the training sample in solving the classification problem.

(a) Monodisperse ion-exchange resins (b) Ordinary ion-exchange resins

Fig. 1. Samples images of ion-exchange resins

Image preprocessing. Image preprocessing using standard computer vision methods includes the following steps:

- Noise reduction, for example, using Gaussian blur, dilation and erosion

- Binarization, which is necessary to unify images for subsequent segmentation

Figure 2 shows the image after applying the Gaussian filter and binarization:

Image segmentation. Segmentation is the separation of areas that are homogeneous according to some criterion, for example, color intensity. In the context of this problem, segmentation is need to localize individual beads of ion-exchange resin in the image.

Fig.. 2. Image after preprocessing

There are two alternative approaches to solving the segmentation problem: - by highlighting the boundaries of the regions;

- by increasing the points of the area.

Segmentation by highlighting the boundaries of the regions

In this step, the boundaries are determined using the Canny detector algorithm. After that using the algorithm Hough transform, the circles will be determined. In figure 3 shown images for that two steps:

(a) Result of Canny detector algorithm (b) Result ofHough transform

Fig.. 3. Segmentation by highlighting the boundaries of the regions Segmentation by increasing the points of the area

In this method used algorithm Marker-based Watershed algorithm. In figure 4 shown results of this method:

(a) Selection by borders (b) Selection by areas

Fig. 4. Results of method Marker-based Watershed

Results. As a result, of the studied segmentation methods for monodispersed ionexchange resins, the methods Canny edge detector and Hough transform were shown the best results with accuracy 80%.

For conventional ion exchange resins, standard computer vision methods didn&t provide an acceptable result. To solve this problem, other approaches are required, for example, the use of convolutional neural networks.

References / Список литературы

1. Water purification by ion exchange. The main characteristics of ion exchangers. [Electronic Resource]. URL: http://www.medianafilter.com.ua/water_filter_ion_exchange.html/ (date of access: 03.06.2020).
ion-exchange resin image segmentation image filtering canny edge detector hough transform watershed with markers ИОНООБМЕННАЯ СМОЛА СЕГМЕНТАЦИЯ ИЗОБРАЖЕНИЙ ФИЛЬТРАЦИЯ ИЗОБРАЖЕНИЙ ДЕТЕКТОР ГРАНИЦ КАННИ
Другие работы в данной теме:
Контакты
Обратная связь
support@uchimsya.com
Учимся
Общая информация
Разделы
Тесты